Merge upstream→gfx11 by eble-amd · Pull Request #1368 · ROCm/vllm

eble-amd · 2026-10-13T19:21:26Z

Purpose

Merge c1b69aa from upstream in two stages.

Manual conflict resolution at 9163190:

accept adjacent insertions
delete stopgap code commented as such

Test Plan & Result (agentic)

AMD Ryzen AI MAX+ 395 (Radeon 8060S, gfx1151), ROCm 10.2.0a20261013, PyTorch 2.13.0, Triton 3.8.0.
Base 680283c57b18, PR 86995aa4fedb; one run per case.

Case	TTFT ms base	TTFT ms PR	TTFT Δ	TPOT ms base	TPOT ms PR	TPOT Δ
Qwen3.6-35B-A3B-AWQ-4bit_VLM	379.3	378.1	−0.3%	31.41	31.41	0.0%
Qwen3.6-35B-A3B-W4A16-llmcompressor-mtpfix_VLM_dyn-lm-int8g32_MTP-mmstar	200.4	200.3	0.0%	10.03	10.03	0.0%
Qwen3.6-35B-A3B-W4A16-llmcompressor_VLM_4ksys_1k_320x240_100out_dyn-lm-int8g32_prefill-dequant	456.9	451.2	−1.2%	12.68	12.61	−0.5%
Gemma-4-E4B-IT_VLM_AWQ-INT4_DFlare-B16-K15_RoboVQA	304.3	302.9	−0.5%	5.97	6.02	+0.8%
Gemma-4-31B-IT_VLM_?_AWQ-4bit_MTP-mmstar	1260.1	1238.1	−1.7%	36.32	36.50	+0.5%
Qwen3-VL-4B-Instruct-AWQ-4bit-lm_head_int8-mmstar	103.0	103.7	+0.8%	11.99	11.99	0.0%

Input and output token counts and speculative-decoding acceptance lengths are identical between base and PR in every case.

Essential Elements of an Effective PR Description Checklist

The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
The test plan, such as providing test command.
The test results, such as pasting the results comparison before and after, or e2e results
(Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
(Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

…project#53780) Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Co-authored-by: OpenAI Codex <codex@openai.com> Co-authored-by: OpenAI Codex <noreply@openai.com>

) Signed-off-by: Andreas Karatzas <akaratza@amd.com>

Signed-off-by: khluu <khluu000@gmail.com> Co-authored-by: Kimi Code <noreply@moonshot.cn> Co-authored-by: Codex <noreply@openai.com>

…the KV block LCM (vllm-project#53007) Signed-off-by: Bill Nell <bnell@redhat.com> Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: OpenAI Codex <codex@openai.com>

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>

…ns (vllm-project#55780) Signed-off-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com>

Signed-off-by: Djordje Ramic <djoramic@amd.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

…llm-project#55223) Signed-off-by: sfeng33 <4florafeng@gmail.com> Signed-off-by: Flora Feng <4florafeng@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>

… was encountered` (vllm-project#55924) Signed-off-by: yewentao256 <zhyanwentao@126.com>

…llm-project#49104) Signed-off-by: cjackal <44624812+cjackal@users.noreply.github.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Co-authored-by: Flora Feng <4florafeng@gmail.com>

Signed-off-by: khluu <khluu000@gmail.com>

Signed-off-by: lcskrishna <lollachaitanya@gmail.com>

Signed-off-by: Shiyang Chen <shiychen@nvidia.com> Signed-off-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Co-authored-by: Isotr0py <Isotr0py@outlook.com>

…5301) Signed-off-by: BANANASJIM <bananasjim1@gmail.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

…V2 (vllm-project#54788) Signed-off-by: Stefano Castagnetta <scastagnetta@nvidia.com> Co-authored-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>

vllm-project#55472) Signed-off-by: Jing Wang <jingwang96@qq.com>

…roject#51925) Signed-off-by: Qidong Su <qidongs@nvidia.com> Signed-off-by: Qidong Su <soodoshll@gmail.com> Co-authored-by: OpenAI Codex <codex@openai.com>

…llm-project#53491) Signed-off-by: Nick Hill <nickhill123@gmail.com>

…53319) Signed-off-by: Michael Lazos <mlazos@meta.com> Co-authored-by: OpenAI Codex <codex@openai.com>

…apper (vllm-project#55954) Signed-off-by: sfeng33 <4florafeng@gmail.com>

vllm-project#55099) Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>

Signed-off-by: omerpaz95 <omerpaz95@gmail.com>

…ed request (vllm-project#54853) Signed-off-by: Zhewen Li <zhewenli@inferact.ai> Co-authored-by: Zhewen Li <zhewenli@inferact.ai> Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com> Co-authored-by: Codex <noreply@openai.com>

Signed-off-by: Djordje Ramic <djoramic@amd.com>

vllm-project#55772) Signed-off-by: Djordje Ramic <djoramic@amd.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>

…anization (vllm-project#55411) Co-authored-by: Codex <noreply@openai.com> Signed-off-by: Bugen Zhao <i@bugenzhao.com>

…e encoding cannot arrive (vllm-project#55290) Signed-off-by: Tianyu Guo <guoty@inferact.ai> Co-authored-by: OpenAI Codex <codex@openai.com>

…llm-project#54033) Signed-off-by: Akine-Ko <2923601304@qq.com>

…llm-project#56447) Signed-off-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2@liverpool.ac.uk>

…hat template (vllm-project#56017) Signed-off-by: 김재억 <gadian88@gmail.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

… pipeline parallelism (vllm-project#53699) Signed-off-by: xiaoming <1259730330@qq.com>

…del_len (vllm-project#56446) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

…gle module (vllm-project#48866) Signed-off-by: Guy Stone <guys@spotify.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

…ect#56459) Signed-off-by: Djordje Ramic <djoramic@amd.com>

…IO (vllm-project#53675) Signed-off-by: Isotr0py <Isotr0py@outlook.com>

…ance improvement for small batch (vllm-project#55356) Signed-off-by: yewentao256 <zhyanwentao@126.com>

Signed-off-by: yewentao256 <zhyanwentao@126.com>

Signed-off-by: Lucain Pouget <lucain@huggingface.co> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

…DCP mode (vllm-project#53664)" (vllm-project#56429) Signed-off-by: Shanshan Shen <87969357+shen-shanshan@users.noreply.github.com>

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Signed-off-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com>

…ss the API key) (vllm-project#56269) Signed-off-by: Sören Dréano <gailenstorm@gmail.com>

vllm-project#46994) Signed-off-by: Chris Eastwood <chris.eastwood@pwn4g3.dev> Signed-off-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

…ct#56160) Signed-off-by: Andy Lo <andy@mistral.ai>

…writes (vllm-project#53280) Signed-off-by: mgoin <mgoin64@gmail.com>

…6499) Signed-off-by: mgoin <mgoin64@gmail.com>

…55426) Signed-off-by: Chengcheng Pei <5881383+chengchengpei@users.noreply.github.com> Co-authored-by: Chengcheng Pei <5881383+chengchengpei@users.noreply.github.com> Co-authored-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>

Signed-off-by: Lyle Lai <lyle.lai@cienet.com>

…oject#56312) Signed-off-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>

…vllm-project#55424) Signed-off-by: Vincent <vincexxchan@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

Manual conflict resolution: - accept adjacent insertions - delete stopgap code commented as such Signed-off-by: Dan Eble <Dan.Eble@amd.com>

No conflicts.

MatthewBonanni and others added 30 commits September 8, 2026 16:01

[2/N][KV Connector][NIXL] Support per-region transfer geometry (vllm-…

e0aaef8

…project#53780) Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Co-authored-by: OpenAI Codex <codex@openai.com> Co-authored-by: OpenAI Codex <noreply@openai.com>

[Bugfix] Handle null RoPE parameters for NoPE layers (vllm-project#55949

18615ad

) Signed-off-by: Andreas Karatzas <akaratza@amd.com>

[CI] Split long misc test groups by command (vllm-project#52346)

808353f

Signed-off-by: khluu <khluu000@gmail.com> Co-authored-by: Kimi Code <noreply@moonshot.cn> Co-authored-by: Codex <noreply@openai.com>

[CI] Reuse ColQwen3 models across pooling tests (vllm-project#55889)

cb22234

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>

[Attention] Require explicit DCP support from attention implementatio…

db3814a

…ns (vllm-project#55780) Signed-off-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com>

[CI][ROCm] Increase timeouts for AMD MI300 jobs (vllm-project#55919)

73f61d1

Signed-off-by: Djordje Ramic <djoramic@amd.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

[Perf] Eliminate full-history reasoning scans for structured outputs (v…

29f46ac

…llm-project#55223) Signed-off-by: sfeng33 <4florafeng@gmail.com> Signed-off-by: Flora Feng <4florafeng@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>

[Perf] Eliminate full-history reasoning scans for structured outputs (v…

0a742da

…llm-project#55223) Signed-off-by: sfeng33 <4florafeng@gmail.com> Signed-off-by: Flora Feng <4florafeng@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>

[Kimi Bug] Fix kda ima `Triton Error [CUDA]: an illegal memory access…

bfb443a

… was encountered` (vllm-project#55924) Signed-off-by: yewentao256 <zhyanwentao@126.com>

[Misc] Bump openai to >=2.25.0 to support namespace tools types (v…

bf33961

…llm-project#49104) Signed-off-by: cjackal <44624812+cjackal@users.noreply.github.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Co-authored-by: Flora Feng <4florafeng@gmail.com>

[CI] Fix ARM64 test dependency builds with GCC 15 (vllm-project#55877)

05e52f6

Signed-off-by: khluu <khluu000@gmail.com>

[ROCm][DI][CI] Enable WideEP Intranode tests (vllm-project#53195)

28a2cce

Signed-off-by: lcskrishna <lollachaitanya@gmail.com>

[Bugfix] Fix OpenPangu multimodal embedding merge (vllm-project#55941)

5af4cc3

Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>

[Bugfix][Spec Decode] Preserve target parallel config (DCP) for DSpark (

c268198

vllm-project#55472) Signed-off-by: Jing Wang <jingwang96@qq.com>

[Kernel] Enable optimized FlashInfer add-RMSNorm NVFP4 fusion (vllm-p…

9e137d8

…roject#51925) Signed-off-by: Qidong Su <qidongs@nvidia.com> Signed-off-by: Qidong Su <soodoshll@gmail.com> Co-authored-by: OpenAI Codex <codex@openai.com>

[Core] Enhance cpu<->gpu sync checking to include paged async copies (v…

4aadb0f

…llm-project#53491) Signed-off-by: Nick Hill <nickhill123@gmail.com>

[Kernel] Add NVFP4 support to the torch linear backend (vllm-project#…

28a73cc

…53319) Signed-off-by: Michael Lazos <mlazos@meta.com> Co-authored-by: OpenAI Codex <codex@openai.com>

[Bugfix] Parse DSML tool calls when the model omits the tool_calls wr…

d98c8c0

…apper (vllm-project#55954) Signed-off-by: sfeng33 <4florafeng@gmail.com>

[ROCm][Perf][Bugfix] Multi-stream perf improvements; rocprofiler fixes (

bc8587f

vllm-project#55099) Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>

[Docs][EC Connector] CPU EC Connector usage Docs (vllm-project#54522)

a97dacb

Signed-off-by: omerpaz95 <omerpaz95@gmail.com>

[CI][IR] Speed up vLLM IR test group (vllm-project#55965)

21a117a

Signed-off-by: Djordje Ramic <djoramic@amd.com>

[Bugfix][Elastic EP] Route only to surviving engines during scale-down (

c6ba5e8

vllm-project#55772) Signed-off-by: Djordje Ramic <djoramic@amd.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>

[Rust Frontend] Simplify reasoning parser initialization and test org…

050f29c

…anization (vllm-project#55411) Co-authored-by: Codex <noreply@openai.com> Signed-off-by: Bugen Zhao <i@bugenzhao.com>

[Bugfix][EC Connector] Fail the request, not the engine, when a remot…

f1aeff6

…e encoding cannot arrive (vllm-project#55290) Signed-off-by: Tianyu Guo <guoty@inferact.ai> Co-authored-by: OpenAI Codex <codex@openai.com>

[Refactor][EC Connector] Add backend extension points to ECCPUWorker (v…

4154163

…llm-project#54033) Signed-off-by: Akine-Ko <2923601304@qq.com>

zRzRzRzRzRzRzR and others added 23 commits September 11, 2026 12:31

[Bugfix] Fix GLM-OCR MTP position masking during CUDA graph capture (v…

89dbb26

…llm-project#56447) Signed-off-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2@liverpool.ac.uk>

[Bugfix][Scoring] Warn when serving original Qwen3 reranker without c…

912dfb3

…hat template (vllm-project#56017) Signed-off-by: 김재억 <gadian88@gmail.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

[Bugfix] Fix Qwen3-VL and Cosmos3-Edge text architectures for CPU and…

a5b714e

… pipeline parallelism (vllm-project#53699) Signed-off-by: xiaoming <1259730330@qq.com>

[Bugfix] Align vLLM YaRN with Transformers and stop re-scaling max_mo…

c191787

…del_len (vllm-project#56446) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

[Metrics] Consolidate Prometheus histogram bucket defaults into a sin…

d5a9d0f

…gle module (vllm-project#48866) Signed-off-by: Guy Stone <guys@spotify.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

[ROCm][Docker] Pin AINIC apt repo to snapshot 1.117.5-a-77 (vllm-proj…

5392fbc

…ect#56459) Signed-off-by: Djordje Ramic <djoramic@amd.com>

[Multimodal] Use GPU NVDEC for EPD encoder-only instance video media …

22258a2

…IO (vllm-project#53675) Signed-off-by: Isotr0py <Isotr0py@outlook.com>

[Kimi Perf] Group fp8 mla cahche insertion, 4~6x kernel level perform…

1e1060f

…ance improvement for small batch (vllm-project#55356) Signed-off-by: yewentao256 <zhyanwentao@126.com>

[Deprecation] Deprecate items scheduled for 0.29 (vllm-project#55353)

5fe77ae

Signed-off-by: yewentao256 <zhyanwentao@126.com>

[httpx migration] Import httpx from huggingface_hub (vllm-project#56460)

b873398

Signed-off-by: Lucain Pouget <lucain@huggingface.co> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Revert "[Rocm][Kimi-k3] Fix pipeline_parallel support for the kimik3 …

127143e

…DCP mode (vllm-project#53664)" (vllm-project#56429) Signed-off-by: Shanshan Shen <87969357+shen-shanshan@users.noreply.github.com>

[Model][ROCm] Enable DeepSeek V4 Vision (vllm-project#55107)

9dd969d

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Signed-off-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com>

[Docs] Correct API key authentication scope (/inference does NOT bypa…

9a35c08

…ss the API key) (vllm-project#56269) Signed-off-by: Sören Dréano <gailenstorm@gmail.com>

[Bugfix][MLA] Read sparse model settings from text config (vllm-proje…

dc07f16

…ct#56160) Signed-off-by: Andy Lo <andy@mistral.ai>

[Kernel][MoE] Optimize batched_moe_align_block_size with cooperative …

8a7f98c

…writes (vllm-project#53280) Signed-off-by: mgoin <mgoin64@gmail.com>

[Agents] Link Triton skill to JIT kernel warmup guide (vllm-project#5…

00e5cda

…6499) Signed-off-by: mgoin <mgoin64@gmail.com>

Upgrade tpu-inference to v0.29.0 (vllm-project#56388)

dffb863

Signed-off-by: Lyle Lai <lyle.lai@cienet.com>

[MRV1] Scope breakable cudagraphs to the piecewise path only (vllm-pr…

cc5dd0a

…oject#56312) Signed-off-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>

[Bugfix][KV Connector] Only enforce disk block alignment for O_DIRECT (…

c1b69aa

…vllm-project#55424) Signed-off-by: Vincent <vincexxchan@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

Merge commit '9163190dda009a310d8c175d63860ac7c671ca90'

9e1c896

Manual conflict resolution: - accept adjacent insertions - delete stopgap code commented as such Signed-off-by: Dan Eble <Dan.Eble@amd.com>

Merge commit 'c1b69aa0d49b88016b5b2156ae61649c2c2df282'

86995aa

No conflicts.

eble-amd requested review from mgehre-amd, roberteg16 and zohaibm-amd October 13, 2026 19:21

eble-amd requested review from AndreasKaratzas and dllehr-amd as code owners October 13, 2026 19:21

mgehre-amd approved these changes Oct 13, 2026

View reviewed changes

eble-amd merged commit 4685ba1 into gfx11 Oct 14, 2026
17 of 19 checks passed

Navigation Menu

Repository navigation

Merge upstream→gfx11 - #1368

Merge upstream→gfx11#1368
eble-amd merged 182 commits into
gfx11from
eble.merge-from-upstream

eble-amd commented Oct 13, 2026 •

edited by github-actions Bot

Loading

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

Conversation

eble-amd commented Oct 13, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Test Plan & Result (agentic)

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

eble-amd commented Oct 13, 2026 •

edited by github-actions Bot

Loading